Future of manufacturing: optimization algorithms
If the “Kaizen” method, traditionally adopted in industrial production, had a mathematical form, this would be that of “optimization algorithms“.
But we have to take a step back to understand the meaning of this statement and especially the connection with the world of industrial production.
Mathematical Optimization is the field where we try to find an optimal solution to a mathematical problem, mainly by maximizing or minimizing mathematical problems.
The fundamental concepts to know about this process are two: constraints and the objective function. Constraints are quantified in a minimum or maximum value, either to be reached or not to be exceeded. The objective function, instead, is not expressed by a reference value but by a direction (toward maximum or minimum) whose ideal end-point is the better result achievable.
The algorithm in mathematics is a method of solving a problem step-by-step. And in computer science is exactly the same: within the software, it is basically a sequence of instructions that allow you to anticipate/solve a problem.
Making a synthesis, optimization algorithms are methods to reduce losses and provide the most accurate solution possible to the problem described by the mathematical model.
Then, is clear why they are natural means of “continuous improvement” for any process/area to which they are applied. And this is true, more than ever, also in the context of industrial production, where today technology and computer science have become mandatory and play a crucial role.
Think, for example, of supply chain management, where production planning must:
- be able to meet market demand (which has become increasingly complex and unpredictable)
- meet a variety of factory constraints such as production line capacity, workforce skills, machine capacity, materials accessibility, overlapping orders, etc.
Without the application of machine learning and deep learning, where optimization algorithm plays a key role, in the Supply Chain wouldn’t be possible to build any model to make predictions.
The software platforms that are implemented must be able to respect the complexity generated by the many variables involved, but at the same time, they must be effective, fast, and intuitive to use.
This is what the d-one platform aims to do, through a project of revamping its internal optimizer in its MES software. This optimizer suggests to the customer one or more possible strategies in order to reach a particular objective of a determined productive process, as an example to try to maximize the profits and/or to maximize the throughput. (Case investigated by the young Simone Galante in his thesis project “Optimal allocation of resources in a production management system.”)
The strength of the d-one platform is that it applies whole linear programming to better manage machine constraints. And it is a significant advantage, having already within the platform these linear programming models that generate the right “suggestions” at the right time.
The currently work in progress ones are:
- Maximize revenue – Suggests a production strategy that maximizes the revenue obtained by the net profit of each.
- Saturate capacity – Suggests a production strategy that maximizes the units of materials produced overall.
- Minimize stock – Suggests a production strategy that minimizes the amount of units kept in the stock while also penalizing any unsatisfied demand.
For the first two, there are the following available constraints:
- Machine availability
- Min/Max amount of material [for specific material(s)]
- Min/Max amount of groups of materials
- Min/Max amount of produced materials by machine [for specific machine(s)]
- Min/Max amount of produced materials by group of machines
- Reach demand at max (Do not produce over demand)
- Batch size
- Max production overall